Azelf
Evidence-led API maintenance: detect breaking changes, propose bounded repairs, and independently verify restored behavior.
A maintenance loop with evidence
Azelf is a maintenance laboratory for applications that depend on changing APIs. It observes a healthy consumer, captures a reproducible failure, gives a repair executor a bounded workspace, and checks the proposed change independently before it reaches review.
The point is not to let an agent guess its way through an outage. It is to make repair constrained, inspectable, and reversible: source identities, failure evidence, candidate diffs, verification results, and release decisions stay connected.
Detect → Repair → Verify
Detect: establish a passing baseline, introduce a controlled upstream change, and confirm the broken capability with repeated observations.
Repair: freeze the evidence and restrict edits to the approved integration surface inside an isolated workspace.
Verify: apply the candidate to a fresh source export and run trusted checks outside the candidate's control, testing restored behavior and preserved behavior separately.
What I built
I designed the control flow, evidence model, repair boundary, independent verifier, release gates, and operator interface. The result is a working TypeScript and React lab that makes the full journey visible—from a healthy permit dashboard to a bounded repair and a reviewable release.